Phoneme processing skills are reflected in children's MMN responses

ABSTRACT Phonological awareness (PA), the core contributor in phoneme processing abilities, has a link to later reading skills in children. However, the associations between PA and neural auditory discrimination are not clear. We used event‐related potential (ERP) methodology and neuropsychological testing to monitor the neurocognitive basis of phonological awareness in typically developing children. We measured 5–6‐year‐old children's (N=70) phoneme processing, word completion and perceptual reasoning skills and compared their test results to their brain responses to phonemic changes, separately for each test. We found that children performing better in Phoneme processing test showed larger mismatch negativity (MMN) responses than children scoring lower in the same test. In contrast, no correspondence between test scores and brain responses was found for Auditory closure. Thus, the results suggest that automatic auditory change detection is linked to phoneme awareness in preschool children. HIGHLIGHTSBehavioural and ERP data were collected from 70 typically developing children.Better phoneme processing skills were associated with larger MMN responses.No correspondence between intelligence measures and MMN or LDN responses was found.


Introduction
Literacy skills are among the most crucial abilities for successful functioning in our society.
Achieving them early on helps a child to do well in school and succeed in later studies.In preschool children, phonological awareness (PA), the ability to perceive and manipulate sounds in spoken language, predicts later reading skills (Kirby et al., 2003;Silvén et al., 2004).Furthermore, the success in tests investigating PA in elementary school children seems to differentiate children with average and above average reading skills (Savage et al., 2005), although it is not clear if the relationship between PA and learning to read is causal or correlational (see e.g., Castles & Coltheart, 2004;Melby-Lervåg et al., 2012).Our knowledge of the correspondence between behavioural measures of pre-reading skills, such as phonological awareness, and neural prerequisites is still incomplete.There is much to learn about how success in neuropsychological tests manifests itself in the developing brain of children.Here we use well-established event-related potential (ERP) methodology to probe the neurocognitive basis of phonological awareness in typically developing children.

Mismatch negativity
Mismatch negativity (MMN) is a negative polarity component of ERPs that is thought to reflect the discrimination of change in a stream of repeating sounds (Näätänen, 1992;Winkler et al., 2009).
According to the current theory, the brain predicts, i.e., forms a neural representation of the regular features in the auditory input and when a change is detected, an MMN response is elicited.MMN appears to originate from two areas: bilaterally supratemporal planes of the auditory cortices and prefrontal cortex (Näätänen & Escera, 2000;Rinne et al., 2000).MMN occurs even when the subject is not attending to the stimuli, and this makes it a practical tool to investigate young children that are easily distracted and sometimes unmotivated to participate experimental tasks (Näätänen et al., 2010; for a review, see e.g.Näätänen et al., 2007).

Late discriminative negativity
n comparison ith N, N seems to ha e distinct neural generators ( eponienè et al., 2004;Hommet et al., 2009), and thus should not be regarded as a late manifestation of the MMN.
Furthermore, unlike MMN, LDN is larger for smaller deviants (Bishop et al., 2011;eponienè et al., 2004).Currently, LDN is thought to reflect additional cognitive processing of subtle changes in auditory stimuli, and not to be linked to attentive or sensory processes in the brain ( ishop et al , 2011eponien et al., 1998;Datta et al., 2010;Shafer et al., 2005).
Some studies have reported LDN to be more pronounced for speech than non-speech sounds (Bishop et al., 2011;Korpilahti et al., 2001;Korpilahti et al., 1996;Kuuluvainen et al., 2016).Yet, the stimulus types in these studies have not always been acoustically comparable (Bishop et al., 2011;Korpilahti et al., 1996) and therefore the reason for differences in ERP amplitudes is not clear.There are also studies with matched stimuli that have not found any differences between linguistic and non-linguistic paradigms ( eponienè et al., 2002;Davids et al., 2011).Overall, the functional significance of LDN response still needs clarification.

Links between neuropsychological measures and neurophysiological indices
Converging evidence shows that auditory ERPs and behavioural discrimination ability correspond in adults (Novitski et al., 2004;Winkler et al., 1999; for reviews, see Kujala & Näätänen, 2010;Kujala et al., 2007) and, according to some studies, in children (Kraus et al., 1996;Maurer et al., 2003).Additionally, association between children's neuroph siological measures and their skills in speech-related tests has come up in several studies (Kujala et al., 2001;Lovio et al., 2010;Lovio et al., 2012;).For example, Lovio et al. (2010) found that 6-year-old children with familial risk for dyslexia both scored worse in phonological test and showed smaller MMNs elicited by speech sound changes than control children.urthermore, some findings support the ie that neuroph siological measures predict outcomes in speech-related tests ( m l inen et al , 201 Jansson-Verkasalo et al., 2004;Kuhl et al., 2008;Maurer et al., 2009).
However, the association between ERPs and behavioural measures is not always straightforward and sometimes children do worse in tests than predicted from their brain responses (Bradlow et al., 1999).Thus, although it seems evident that there is a correspondence between ERP measures and linguistic test scores, the issue still needs clarification.
The relationships between intelligence measures and ERPs to auditory stimuli are largely understudied.Alternatively, the scarce literature may depend on the publication bias, since studies not finding any link between investigated measures tend not to be published.Most research seems to focus on schizophrenic (Kawakubo et al., 2006;Light & Braff, 2005a, 2005b) or autistic patients (Weismüller et al., 2015).However, Light et al. (2007) found that the MMN response of healthy adults to duration change correlated ith participants' o erall le el of functional status, as measured by Global Assessment of Functioning Scale (Hall, 1995).As for the children, Mikkola et al. (2007) found a correlation between MMN amplitudes to frequency changes and verbal IQ and verbal fluency test results when studying preterm and full-term children at the age of five years.In addition, when comparing children with speech disorders to typically developing children, Bauer et al. (2009) found that the amplitude of MMN correlated with the auditory memory span test results.
Some studies have focused on typically developing children.Partanen et al. (2013a) discovered a connection between MMN amplitudes for intensity changes and erbal Q tests in −12-year-old children.In addition, Liu et al. (2007) reported that the peak amplitudes of MMN and LDN responses of highl intelligent 11−12-year-old children for consonant change were larger and the LDN latency was shorter than those of their peers of average intelligence.As most of the studies show evidence for differences between healthy adults and groups of special features (e.g., schizophrenia patients), there is still much to be learned about associations between ERPs and intelligence measures within subject groups with no clinical background.
In our study we tested seventy typically developing children with three different tests, and thus aimed to find out whether subtle differences in neuropsychological test performance would be reflected in brain responses for phonemic stimuli.Finding such differences would suggest that there are fine-tuned links between neural substrates and testable linguistic or other cognitive abilities.Our hypothesis was that children with higher scores in linguistic tests would show larger MMN and LDN responses than the children with lower scores in the same tests.We also hypothesized that children having higher scores in tests for intelligence would show larger MMN amplitudes than their lower scoring peers.If our hypotheses prove right, it would mean that in typically developing children there is a direct link between phonological awareness and/or intelligence and hearing subtle details in linguistic sounds.Furthermore, if there are differences in how discriminating different phoneme change types differentiate children with higher and lower scores in each test, we will learn more about which sound features are more closely linked to phonological awareness, or intelligence, than others.

Participants
All 75 participants ere 5−6-year old kindergarten children (mean age 5 years 9 months).Due to less than 65% accepted stimulus trials in EEG data 5 participants were excluded, and data from 70 (44 female) participants were left for further analyses.The children attended 12 different municipal Finnish language kindergartens in Espoo region, and 57 of them were native Finnish speakers.The rest were bilingual, and spoke Russian (3), Estonian (2), Albanian (4), Somali (2), Swedish (1) and Armenian (1) as their native language.Bilingual children all spoke and understood Finnish at least relatively well.Among the children were thirteen whose parents reported their children either having language problems or having close relatives with dyslexia.However, there were no official diagnoses and these children's test results [Phoneme Processing: t(68) = .327,p = .745;Auditory closure: t(68) = -.614,p = .542;Perceptual Reasoning Index: t(68) =1.522, p = .133]did not differ from those of the other children in the sample (for children's indi idual scores, see Appendix, Table 1).Furthermore, we conducted two rANOVAs comparing the mean amplitudes of MMN and LDN responses of children with possible language problems and the other children on nine chosen electrodes for all deviants.As we found no differences [MMN: F(1, 68) = 1.251, p = .283;LDN: F(1, 68) = .500,p = .482],we included these children in the experiment.
The parents or guardians signed a written informed consent and the children gave their verbal assent before the experiment.The experiment protocol was approved by The Ethical Committee of the Humanities and Social and Behavioural sciences in the University of Helsinki, Finland.

Neurocognitive assessments
Children were tested with Phoneme processing subtest (NEPSY II, Korkman, Kirk & Kemp, 2008), Auditory closure subtest (ITPA, Kirk et al., 1972), along with Block design and Matrix reasoning subtests (WISC-IV, Wechsler, 2010).All the tests were rehearsed according to the test guidelines with children before the experiment, and they were well understood also by the bilingual children.
Phoneme processing (PP) subtest measures phonological awareness and auditory memory.In the first section, the child sees three pictures and hears names for each object in the pictures.The experimenter pronounces then a phoneme combination that is included in one of the object names and the child has to point to the object, which the syllable belongs to.In the next section the child is asked to remove a phoneme or combination of phonemes from the uttered word and say the resulting ord out loud (" a /tak:a/, 'fireplace' Then sa the same ord ithout /t/".The right ans er is /ak:a/, 'an old oman') n the final section, (if it is reached) the child has to replace a phoneme or combination of phonemes ith gi en phonemes (" a /helmi/, 'pearl' No sa the same ord but replace /i/ ith /a/" The right ans er is /helma/, 'hem of skirt') Testing is stopped after six consecutive wrong answers.Reasoning Index (PRI) as an indication of non-verbal intelligence.

The ERP paradigm
The stimuli were made with semisynthetic Speech Generation Method (for details, see Alku, Tiitinen & Näätänen, 1999) and were presented in a multifeature paradigm (Figure 1) (Näätänen et al., 2004).Unlike in traditional oddball paradigm where the deviant stimulus occurs typically 10-20% of the time, in multifeature paradigm every other stimulus is a standard and every other a deviant.In one paradigm there need to be several different deviant types that alternate so that every deviant type differs from the standard in only one feature (e.g.frequency or intensity).Thus, even though the deviants occur in 50% of the heard sounds, each deviant type appears only in e.g., 10% of the stimuli.It has been shown that MMN responses of healthy adults in multifeature paradigm are similar to those elicited by traditional oddball paradigm (Pakarinen et al., 2009;Kujala, Lovio, Lepistö, Laasonen & Näätänen, 2006).In addition, in six-year-old children's responses to five different deviant types recorded in a multifeature paradigm were comparable to those recorded in an oddball paradigm (Lovio et al., 2009).The multifeature paradigm is a very efficient way to collect a large amount of data in short time, and thus it is very convenient method to measure children.In our study, one deviant type appeared approximately in 10% of the stimuli and the actual measurements took less than 20 minutes.Had we recorded the same amount of deviant trials in an oddball paradigm with 15% of the stimuli being deviants, it would have taken more than 50 minutes to measure one subject.For a child of 5-6 years of age it is a long time to sit still.Furthermore, as the previous literature is not coherent about the phoneme change types that might or might not be linked to test performance, we wanted to include different phoneme changes in our study.
The deviant stimuli were consonant change CON (P = .10),vowel change VOW (P = .10),vowel duration change DUR (P = .10),intensity change INT (louder P = .05and softer P = .05)and frequency change FRE (higher P = .05,lower P = .05).The duration of the stimuli was 170 ms, excluding deviant DUR (100 ms).Onsets of the stimuli were 500 ms apart from each other.F0 was 101 Hz for all other stimuli expect for the FRE deviants which had the f0s of 93 Hz and 109 Hz.
Intensity of the stimuli was ~70 dB (SPL) excluding the intensity deviants that were 63 dB and 77 dB.There were 465 stimuli in each of the four blocks and each block lasted for about 5 minutes.
The order of the blocks was counterbalanced and the total EEG recording net time was 20 minutes.
The identical experiment paradigm has been used successfully in measuring MMN responses in both adults (Pakarinen et al., 2009) and children (Lovio et al., 2009).
Table 1 The stimuli of the multifeature paradigm.Four blocks (two blocks for each standard stimuli) were played for the participant.The blocks were in randomized order.

The procedure
The EEG measurements and the neurocognitive tests were conducted in the kindergarten premises.
The measurement rooms were as silent as possible, and only the participant and the experimenter(s) ere present EEG measurements and neurocogniti e tests ere conducted during the children's normal daily stay at kindergarten, on separate days.While preparing for and measuring EEG, the child watched an animated movie that was muted during measuring.The participants were asked to avoid unnecessary movement, to ignore the experimental stimuli, and to concentrate on the movie.
The stimuli were presented via headphones (Sony Professional MDR-7506) with short breaks between blocks.Cookies and soft drink was offered during breaks.Preparation, measurement and removing the cap took approximately an hour, as did the neurocognitive testing.

Data recording and processing
The stimuli were presented with Presentation 17.0 (Neurobehavioral Systems, Inc., CA, US) and the EEG was recorded with 32 Ag-AgCl scalp electrodes according to international 10-20 system (ActiCap; Brain Products, Germany).The EEG equipment was portable (Brainvision QuickAmp amplifier).The data were registered with sample rate of 500 Hz and recording reference was the average signal of all electrodes.Additional two active electrodes were placed on the mastoids behind both ears.
Noisy electrodes were interpolated and eye blink artefacts were removed using semi-automatic Besa PCA method.Frequencies under 0.5 Hz and over 30 Hz were filtered out offline and the data were re-referenced to the mean of the mastoids.EEG was epoched from -100 ms before onset to 500 ms after the onset of the stimuli and the epochs including amplitudes exceeding ±120 μV ere excluded from the analyses.The responses were averaged for each participant and the averaged responses were then exported to MATLAB R2012 (The MathWorks Inc., MA, US).
The response for intensity deviant was created by averaging the responses to both intensity changes (louder and softer).In the similar way, the response for frequency deviant was created by

Statistical analyses
A three-way repeated measures ANOVA (5 deviants x 3 front-back electrode lines x 3 left-right electrode lines) was conducted with SPSS 24 (IBM Corporation, NY, USA) with three test scores (PP, AC and PRI) used as covariates and native language as between-subjects factor.Whenever sphericity could not be assumed in the analyses, Greenhouse-Geisser corrections were applied.

Test performance
The average test results for the test scores (PP, AC, PRI) are outlined in Table 2.   3).No other significant main effects of interactions were found for the MMN.Furthermore, no significant main effects or interactions were found on LDN time window.

Phoneme processing
As we hypothesized, children achieving higher scores in the Phoneme Processing subtest of NEPSY II showed larger MMN amplitudes than lower performers.n other ords, the children's performance in the active behavioural test was reflected in the brain responses that were measured in a passive condition.Thus, it seems that automatic auditory change detection has a link, or that it even contributes, to phonological awareness.The mechanism behind this link is subject to speculation.It could be that the change detection reflects the accuracy of the memory trace, and the more accurate it is, the easier it is to make conscious inferences and manipulations of phonemes.
More research is needed to find out if the differences in automatic change detection system are innate or a consequence of different auditory environments that children live in.
It is especially interesting that we were not comparing typically developing children to a clinical population, as most studies have done (e.g., Kujala, 2007;Lovio et al., 2010), but the difference was seen among typically developing children.Therefore, the salience or non-salience of responses is not dependent only on the possible diagnoses in children's linguistic de elopment, but differentiates also typically developing children from each other.Since phonological awareness is a known index of later reading skills, brain responses connected to PA might also tell us early on how the reading skills of a child are likely to develop.
Contrary to what we hypothesized, the children's N responses did not sho differences that might be linked to their performance in phoneme processing.Thus, we found no evidence that would link the LDN response to language processing skills.Further studies are needed to assess the significance of this response that is found in children and is likely to diminish with age (Bishop et al., 2011;Hommet et al., 2009;Liu et al., 2014).

Auditory closure
No differences linked to Auditory closure test emerged in the children's N or N responses.
As we did find significant differences in responses based in subjects' scores of Phoneme processing test, it seems that AC is not as strongly linked to neural sound discrimination abilities as PP.Even though the manipulation of phoneme combinations is required also in this test, AC may be more closely associated with vocabulary size and linguistic memory than with phonological awareness.

Perceptual reasoning skills
Contrary to our second hypothesis, the higher performing children in Perceptual Reasoning Index

Summary and conclusions
The current stud in estigated the associations bet een children's performance in linguistic and cognitive tests and their neural phoneme discrimination skills.We found that children's scores in Phoneme processing test were associated with their neural auditory discrimination abilities.The children scoring higher in the test had larger MMN amplitudes to phonemic changes.Furthermore, we found that children scoring lower in Perceptual reasoning index tests showed right-side dominance in their MMN responses.Auditory closure was not correlated with neural indices in this paradigm.
Taken together, based on ERP measurements and neuropsychological testing of seventy typically developing children, it seems that automatic auditory detection is linked to phonological awareness in preschool children.Further studies are needed to investigate whether this link is correlational or causal, and whether some day it will be possible to predict individual reading skills from auditory discrimination abilities.If so, it would enable one to tailor individual training programs to improve phonological awareness prior the child is supposed to learn to read.This approach would prevent or at least minimize also the problems often occurring in parallel with slow reading skill acquisition, namely low self-esteem and social disintegration.
Table 1 Individual scores for all the tests.LP stands for children whose parents reported them having language problems or their close relatives having dyslexia and NLP for children whose parents did not report any problems.
averaging the responses to increments and decrements of frequency.The difference signals were created for each deviant stimulus by subtracting the average standard response from the average deviant responses for each participant.The standard and deviant trials from all four blocks were combined according to the stimulus category.The electrodes F3, Fz, F4, C3, Cz, C4, P3, Pz and P4 were chosen for further inspection.This is a typical choice in MMN paradigms, as it reveals frontback and left-right distribution of the brain responses.Mean amplitudes of MMN were calculated from the deviant-standard difference signal o er 200−250 ms for CON, FRE, INT and VOW deviants and o er 225−275 ms for DUR deviant.LDN mean amplitudes were calculated over 75− 25 ms for all the de iants The children's N and N responses t picall occur in these time windows and the visual inspection showed that all the inspected responses were prominent within these time ranges.

Figure 2
Figure2shows the standard and deviant responses for the combined conditions for all participants.
differing from the baseline are marked with asterisks.*** p<.001As hypothesized, the test scores in Phoneme processing had a significant main effect on the responses [F(1)= 4.315 ; p=.042] indicating that children with higher scores on the test showed a larger MMN amplitude relative to children with lower scores (for illustrational purposes, see Appendix, Figure 1 that depicts subtraction curves for higher and lower scoring children for all three tests).Main effects of Auditory closure [F(1)=.508; p=.479] or Perceptual reasoning index [F(1)= 1.700 ; p=.197] test scores on the responses were not significant, and neither was subjects' native language [F(1)= 056; p=.308].For MMN, the only significant interaction was Left-Right x PRI [F(2, 130)=3.271;p=.041].The post hoc comparisons were conducted by comparing the estimated mean amplitudes over the left, centre and right electrode lines when PRI scores were set at the first or the third quartile.These comparisons indicated that children with higher scores did not show any lateralization of responses [with higher quartile scores (35): Left: Mean = -2 1μV Central: ean= -2 0 μV Right: ean=-2 1 μV; significance of differences between Left-Right lines: Left vs. Central p=.596,Left vs. Right p=.962, Central vs. Right p=.581], whereas children with lower scores showed significant rightside dominance of the responses [with lower quartile scores (24): Left: Mean = -1.6μV;Central: Mean= -1.6 μV Right: ean= -1.9 μV; significance of differences between Left-Right lines: Left vs. Central p=.726,Left vs. Right p=.042, Central vs. Right p=.026].
test did not show larger MMN or LDN responses to phonemic stimuli than lower performing children.Thus, based on our study, it seems that perceptual measures of intelligence do not strongly correlate with discrimination of phonemic changes.Previous findings have not been very consistent as to which auditory change responses correlate with which intelligence test results; the studies have used different ERP paradigms and different tests.E.g., Light et al. (2007) found a correspondence bet een adults' N responses to duration change and o erall le el of functional status (Global Assessment of Functioning Scale, Hall, 1995).Both Partanen et al. (2013a) and Mikkola et al. (2007) found a link between verbal intelligence scores (WISC-IV, Wechsler, 2010; WPPSI-R, Wechsler, 1995) and responses for sound changes in children, Partanen et al. for intensity change and Mikkola et al. for frequency change.Nevertheless, the studied subjects were not comparable to each other or to the sample of the present study: Partanen et al. studied children representing very wide age distribution and Mikkola et al. compared children born preterm and full-term.As for us, we used non-verbal intelligence scores to measure children's intelligence, and phonemes as stimuli Thus, it seems that the accuracy of auditory memory trace does not have a link to perceptual reasoning based on visual cues.However, we found an interaction suggesting that whereas the responses of children with better perceptual reasoning skills are distributed evenly across the scalp, the responses of children not performing as well in these skills have a right side dominance.It might reflect differences in maturationthe brains of children with better perceptual reasoning skills are activated more evenly by auditory stimuli, while children with less strong performance in the same skills do show more lateralized activation.Yet, the reason for this rightward dominance is unclear.Bauer et al. (2009) found that when looking at area under the curve (AUC) of MMN, children with central auditory processing disorder (CAPD) who scored significantly lower in auditory memory span test than their peers, tended to have left-hemisphere dominance in their MMN.The control children's responses were evenly distributed.Partanen et al. (2013a) found that pre-schoolers had larger MMN responses on left and central electrodes to frequency deviant than school children, and according to Everts et al. (2009) right-lateralization of visuo-spatial processing increases by age, being more evenly distributed in young children than in older adolescents.As the literature of the correspondence between intelligence measures and lateralization of brain responses is scarce and very inconsistent, more research is needed to shed light on the associations between different measures of intelligence and neurophysiological indices.

Figure 1
Figure 1 Averaged subtraction waveforms at Fz electrode for higher and lower scoring groups in (Hannus et al., 2013) test probes if a child is able to produce a complete word out of an incomplete one.It is used regularly by speech therapists in Finland and other Scandinavian countries and has been found to differentiate children with specific language impairment (SLI) from typically developing children(Hannus et al., 2013).Because of its clinical relevance, we were interested to see if AC has a correspondence with neurophysiological measures.The procedure of the test goes as follows: experimenter says an utterance that is incomplete as a word.Participant then has to supplement it with phonemes to make it a proper word.More than one right answer is accepted ("What ord do mean ith /a ai/?" Right ans er might be /a ain/, 'ke ' or /a aimet/, 'ke s')Block design and Matrix reasoning are both subtests of The Perceptual Reasoning Index's section of WISC-IV.Block design (BD) measures visuospatial skills.Children form patterns with red-and-white blocks according to a displayed model, and this performance is timed.In Matrix reasoning (MR) test children are shown an array of pictures with one missing square, and have to select from five options the picture that fits the array.This test measures fluid reasoning skills.In our study, we combined BD and MR according to instructions in WISC-IV to form Perceptual

Table 2
Mean scores for all the tests.N=70, mean age in months (SD) 69.30 (3.1).

Table 3
for all the other deviants) andamplitudes for all children at Fz.Standard deviations in brackets.